Efficient image denoising method based on mathematical morphology reconstruction and the Non-Local Means filter for the MRI of the head

Abstract

The efficient Non-Local Means denoising algorithm modifies the intensity of each pixel by the weighted average of all similar pixels in the noisy image. It stems from the assumption that there are many similar structures in natural images. Many adaptations of the NLM filter has been widely used for MRI image denoising. The Unbiased NLM is a popular one of these methods which subtracts the rician noise bias from the NLM filtered image. The bias can be estimated from the MRI image background. Prior to that, the background needs to be extracted from the image. However, the estimated rician noise bias depends strongly on the segmentation process which affects the algorithm performance. In this work, we propose an accurate segmentation based on morphological reconstruction to separate the image into two regions-foreground and background. Initially, we propose a dynamic structuring element which the shape adapt according to the input image to avoid the problem of choosing an appropriate structuring element. The obtained background is used to estimate the noise bias while the Unbiased NLM filter is applied topically on the obtained foreground using the estimated bias. Experimental results show that the proposed method perform better than the NLM filter and the UNLM under all tested noise levels.